|Publication number||US7725144 B2|
|Application number||US 10/742,110|
|Publication date||May 25, 2010|
|Filing date||Dec 19, 2003|
|Priority date||Apr 4, 2002|
|Also published as||US20050090750|
|Publication number||10742110, 742110, US 7725144 B2, US 7725144B2, US-B2-7725144, US7725144 B2, US7725144B2|
|Inventors||Marwood Neal Ediger, Craig M. Gardner, Edward L. Hull|
|Original Assignee||Veralight, Inc.|
|Export Citation||BiBTeX, EndNote, RefMan|
|Patent Citations (42), Referenced by (2), Classifications (29), Legal Events (4)|
|External Links: USPTO, USPTO Assignment, Espacenet|
This application claims priority under 35 U.S.C §120 as a continuation in part of U.S. patent application Ser. No. 10/116,272, entitled “Apparatus And Method For Spectroscopic Analysis Of Tissue To Detect Diabetes In An Individual,” filed Apr. 4, 2002, now U.S. Pat. No. 7,043,288, issued May 9, 2006, incorporated herein by reference. This application claims priority under 35 U.S.C §119 to U.S. provisional application 60/515,343, Determination of a Measure of a Glycation End-Product or Disease State Using Tissue Fluorescence,” filed Oct. 28, 2003, and U.S. provisional application 60/517,418, filed Nov. 4, 2003 Apparatus and Method for Spectroscopic Analysis of Tissue to Determine Glycation End-products incorporated herein by reference.
The present invention generally relates to determination of disease state from Raman spectroscopy in tissue. More specifically, the present invention relates to methods and apparatuses for determining models that relate Raman spectra of tissue to disease state, and for determining properties of Raman spectra in tissue, and for determination of disease state from Raman spectral properties and from appropriate models.
Diabetes mellitus is a major health problem in the United States and throughout the world's developed and developing nations. In 2002, the American Diabetes Association (ADA) estimated that 18.2 million Americans—fully 6.4% of the citizenry—were afflicted with some form of diabetes. Of these, 90-95% suffered from type 2 diabetes, and 35%, or about 6 million individuals, were undiagnosed. See ADA Report, Diabetes Care, 2003. The World Health Organization (WHO) estimates that 150 million people worldwide suffer from diabetes; type 2 diabetes also represents 90% of all diagnoses worldwide. Unfortunately, projections indicate that this grim situation will worsen in the next two decades. The WHO forecasts that the total number of diabetics will double before the year 2025. Similarly, the ADA estimates that by 2020, 8.0% of the US population, some 20 million individuals, will have contracted the disease. Assuming rates of detection remain static, this portends that, in less than twenty years, three of every 100 Americans will be ‘silent’ diabetics. It is no surprise that many have characterized the worldwide outbreak of diabetes as epidemic.
Diabetes has a significant impact on individual health and the national economy. U.S. health care costs related to diabetes exceeded $132 billion in 2002. Due to the numerous complications that result from chronic hyperglycemia, these costs were distributed over a wide array of health services. For example, between 5 and 10 percent of all U.S. expenditures in the areas of cardiovascular disease, kidney disease, endocrine and metabolic complications, and ophthalmic disorders were attributable to diabetes. See ADA Report, Diabetes Care, 2003. These economic and health burdens belie the fact that most diabetes-related complications are preventable. The landmark Diabetes Control and Complications Trial (DCCT) established that a strict regimen of glucose monitoring, exercise, proper diet, and insulin Research Group, N Eng J Med, 1993. Furthermore, the ongoing Diabetes Prevention Program (DPP) has already demonstrated that individuals at risk for diabetes can significantly reduce their chances of contracting the disease by implementing lifestyle changes such a weight loss and increased physical activity. See DPP Research Group, N Eng J Med, 2002. ADA has recommended that health care providers begin screening of individuals with one or more disease risk factors, observing: “If the DPP demonstrates a reduction in the incidence of type 2 diabetes as a result of one or more of the [tested] interventions, then more widespread screening . . . may be justified”. See ADA Position Statement, Diabetes Care, 2003.
The Fasting Plasma Glucose (FPG) test is one of two accepted clinical standards for the diagnosis of or screening for diabetes. See ADA Committee Report, Diabetes Care, 2003. The FPG test is a carbohydrate metabolism test that measures plasma glucose levels after a 12-14 hour fast. Fasting stimulates the release of the hormone glucagon, which in turn raises plasma glucose levels. In non-diabetic individuals, the body will produce and process insulin to counteract the rise in glucose levels. In diabetic individuals, plasma glucose levels remain elevated. The ADA recommends that the FPG test be administered in the morning because afternoon tests tend to produce lower readings. In most healthy individuals, FPG levels will fall between 70 and 100 mg/dl. Medications, exercise, and recent illnesses can impact the results of this test, so an appropriate medical history should be taken before it is performed. FPG levels of 126 mg/dl or higher indicate a need for a subsequent retest. If the same levels are reached during the retest, a diagnosis of diabetes mellitus is typically rendered. Results that measure only slightly above the normal range may require further testing, including the Oral Glucose Tolerance Test (OGTT) or a postprandial plasma glucose test, to confirm a diabetes diagnosis. Other conditions that can cause an elevated result include pancreatitis, Cushing's syndrome, liver or kidney disease, eclampsia, and other acute illnesses such as sepsis or myocardial infarction.
Because it is easier to perform and more convenient for patients, the FPG test is strongly recommended by the ADA and is in more widespread use than the other accepted diagnostic standard, the OGTT. The OGTT is the clinical gold standard for diagnosis of diabetes despite various drawbacks. After presenting in a fasting state, the patient is administered an oral dose of glucose solution (75 to 100 grams of dextrose) which typically causes blood glucose levels to rise in the first hour and return to baseline within three hours as the body produces insulin to normalize glucose levels. Blood glucose levels may be measured four to five times over a 3-hour OGTT administration. On average, levels typically peak at 160-180 mg/dl from 30 minutes to 1 hour after administration of the oral glucose dose, and then return to fasting levels of 140 mg/dl or less within two to three hours. Factors such as age, weight, and race can influence results, as can recent illnesses and certain medications. For example, older individuals will have an upper limit increase of 1 mg/dl in glucose tolerance for every year over age 50. Current ADA guidelines dictate a diagnosis of diabetes if the two-hour post-load blood glucose value is greater than 200 mg/dl on two separate OGTTs administered on different days.
In addition to these diagnostic criteria, the ADA also recognizes two ‘pre-diabetic’ conditions reflecting deviations from euglycemia that, while abnormal, are considered insufficient to merit a diagnosis of diabetes mellitus. An individual is said to have ‘Impaired Fasting Glucose’ (IFG) when a single FPG test falls between 110 and 126 mg/dl. Similarly, when the OGTT yields 2-hour post-load glucose values between 140 and 200 mg/dl, a diagnosis of ‘Impaired Glucose Tolerance’ (IGT) is typically rendered. Both of these conditions are considered risk factors for diabetes, and IFG/IGT were used as entrance criteria in the Diabetes Prevention Program. IFG/IGT are also associated with increased risk of cardiovascular disease.
The need for pre-test fasting, invasive blood draws, and repeat testing on multiple days combine to make the OGTT and FPG tests inconvenient for the patient and expensive to administer. In addition, the diagnostic accuracy of these tests leaves significant room for improvement. See, e.g., M. P. Stern, et al., Ann Intern Med, 2002, and J. S. Yudkin et al., BMJ, 1990. Various attempts have been made in the past to avoid the disadvantages of the FPG and OGTT in diabetes screening. For example, risk assessments based on patient history and paper-and-pencil tests have been attempted, but such techniques have typically resulted in lackluster diagnostic accuracy. In addition, the use of glycated hemoglobin (HbA1c) has been suggested for diabetes screening. However, because HbA1c is an indicator of average glycemia over a period of several weeks, its inherent variability combines with the experimental uncertainty associated with currently-available HbA1c assays to make it a rather poor indicator of diabetes. See ADA Committee Report, Diabetes Care, 2003. HbA1c levels of diabetics can overlap those of nondiabetics, making HbA1c problematic as a screening test. A reliable, convenient, and cost-effective means to screen for diabetes mellitus is needed. Also, a reliable, convenient, and cost-effective means for measuring effects of diabetes could help in treating the disease and avoiding complications from the disease.
U.S. Pat. No. 5,553,616 (Ham) discloses instruments and methods for noninvasive tissue glucose level monitoring via Raman spectroscopy and spectral processing by neural networks and fuzzy logic. Ham does not describe measurement of any other tissue property, or any method of screening for or monitoring diabetes.
U.S. Pat. No. 5,582,168 (Samuels) discloses apparatus and methods for measuring characteristics of biological tissues and similar materials. These apparatus and methods are described with respect to measurements of the human eye. In addition, the correction methodologies described by these inventors involve only measurements of the elastically scattered excitation light. Samuels describes a simple linear correction technique. Samuels does not disclose noninvasive measurements that allow determination of tissue disease status.
U.S. Pat. No. 5,882,301 (Yoshida) discloses methods and apparatus for obtaining Raman emission from intraocular substances including advanced glycated endproducts (AGEs). Yoshida does not describe a technique for assessing AGEs in skin or for quantifying the AGE concentration as a means to determine disease status.
U.S. Pat. No. 6,044,285 (Chaiken) discloses a system based upon Raman spectroscopy for measuring blood glucose. The described technique relies upon an absorbing species such as hemoglobin acting as a temperature probe. Chaiken does not disclose measurement of advanced glycation endproducts or other analytes relating to disease status. In addition, Chaiken does not describe methods for correction techniques to compensate for local skin absorption or scattering.
U.S. Pat. No. 6,167,290 (Yang) discloses a Raman spectroscopy system for noninvasively measuring blood glucose. Yang does not disclose measurement of advanced glycation endproducts or other analytes relating to screening for or monitoring diabetes status. Furthermore, Yang does not describe methods for correction techniques to compensate for local skin absorption or scattering in order to recover the intrinsic Raman emission signal.
U.S. Pat. No. 6,289,230 (Chaiken) describes an apparatus for the non-invasive quantification of glucose via Raman spectroscopy. Chaiken does not disclose measurement of advanced glycation endproducts or other analytes relating to disease status. In addition, Chaiken does not describe methods for correction techniques to compensate for local skin absorption or scattering.
U.S. Pat. No. 6,352,502 (Chaiken) describes an apparatus based upon Raman spectroscopy for the noninvasive characterization of skin and detection of skin abnormalities. Chaiken does not disclose measurement of advanced glycation endproducts or other analytes relating to diabetes status. Chaiken does not describe methods to extract the intrinsic Raman emission from the detected signal nor multivariate techniques to quantitatively predict analyte concentration.
U.S. Pat. No. 6,560,478 (Alfano) describes an apparatus based upon Raman spectroscopy for examining biological materials. Alfano discloses that the technique can be applied for the diagnosis of disease by measuring characteristic Raman emission associated with blood glucose and other constituents. Alfano does not describe a method or technique for quantifying Advanced Glycation Endproducts as a metric for assess diabetes status. Also, Alfano does not disclose algorithms or methods for recovering intrinsic Raman emission or other techniques to compensate for local tissue variations.
The present invention provides a method of determining disease state in an individual. A portion of the tissue of the individual is illuminated with excitation light, then light returned by the tissue due to inelastic Raman scattering by chemicals within the tissue responsive to the excitation light is detected. The detected light can be combined with a model relating Raman spectra with disease state to determine a disease state of the individual. The invention can comprise single wavelength excitation light, scanning of excitation light (illuminating the tissue at a plurality of wavelengths), detection at a single wavelength, scanning of detection wavelengths (detecting emitted light at a plurality of wavelengths), and combinations thereof. The invention also can comprise correction techniques that reduce determination errors due to detection of light other than that from Raman scattering of a chemical in the tissue. For example, the reflectance of the tissue can lead to errors if appropriate correction is not employed. The invention can also comprise a variety of models relating Raman spectra to disease state, including a variety of methods for generating such models. Other biologic information can be used in combination with the Raman spectral properties to aid in the determination of disease state, for example age of the individual, height of the individual, weight of the individual, history of disease in the individual's family, ethnicity, skin melanin content, or a combination thereof. Fluorescence or near-infrared spectroscopic examination can also be used to supply additional information, for example like that discussed in U.S. patent application Ser. No. 10/116,272, entitled “Apparatus And Method For Spectroscopic Analysis Of Tissue To Detect Diabetes In An Individual,” filed Apr. 4, 2002. The invention also comprises apparatuses suitable for carrying out the method, including appropriate light sources, detectors, and models (for example, implemented on computers) used to relate detected Raman scattering and disease state.
As used herein, “determining a disease state” includes determining the presence or likelihood of diabetes; the degree of progression of diabetes; a change in the presence, likelihood, or progression of diabetes; a probability of having, not having, developing, or not developing diabetes; the presence, absence, progression, or likelihood of complications from diabetes. “Diabetes” includes a number of blood glucose regulation conditions, including Type I, Type II, and gestational diabetes, other types of diabetes as recognized by the American Diabetes Association (See ADA Committee Report, Diabetes Care, 2003), hyperglycemia, impaired fasting glucose, impaired glucose tolerance, and pre-diabetes. “Tissue reflectance characteristic” includes any reflectance property of tissue that is useful in correction of detected light, including as examples the tissue reflectance at the Raman excitation wavelength, the tissue reflectance at the Raman scattering wavelength, and the tissue reflectance at other wavelengths found useful for estimating the tissue's intrinsic Raman scattering spectrum. A “measure of chemical change due to glycemic control” means any change in the chemical characteristics of tissue that is due to glycemic control, examples including concentration, measurements of the presence, concentration, or change in concentration of glycation end-products in tissue; measurements of the rate or change in the rate of the accumulation of such end-products; measurements of tissue membrane thickness or the change, rate of change, or direction of change of such thickness; tissue properties such as tensile strength, strain, or compressibility, or the change, rate of change, or direction of change of such property. A “measure of glycation end-product” means any measure of the presence, time, extent, or state of tissue associated with hyperglycemia, including, as examples, measurements of the presence, concentration, or change in concentration of glycation end-products in tissue; measurements of the rate or change in the rate of the accumulation of such end-products; measurements of the presence, intensity, or change in intensity of Raman scattering signal at Stokes shifts known to be associated with tissue glycation end-products; and measurements of the rate or change in the rate of the accumulation of such Raman signal. When light is described as having a “single wavelength,” it is understood that the light can actually comprise light at a plurality of wavelengths, but that a significant portion of the energy in the light is transmitted at a single wavelength or at a range of wavelengths near a single wavelength.
The drawings, which are not necessarily to scale, depict illustrative embodiments and are not intended to limit the scope of the invention.
Exposure of proteins to glucose generally leads to nonenzymatic glycation and glycoxidation, a process known as the Maillard reaction. The stable endproducts of the Maillard reaction are collectively denoted Advanced Glycation Endproducts (AGEs). In the absence of significant clearance, these AGEs accumulate at rates proportional to the average level of glycemia. The Maillard reaction can be viewed as an aging process that occurs routinely in health and at an accelerated rate in diabetics due to the presence of chronic hyperglycemia. In skin, collagen is the most abundant protein and readily undergoes glycation. Skin collagen AGEs commonly take the form of crosslinks and adducts; pentosidine (a crosslink) and carboxymethyl-lysine (CML, an adduct) are two well-studied examples of skin-collagen AGEs. Other examples of AGEs include fluorolink, pyrraline, crosslines, Nε . . . -(2-carboxyethyl) lysine (CEL) glyoxal-lysine dimer (GOLD), methylglyoxal-lysine dimer (MOLD), 3DG-ARG imidazolone, vesperlysines A, B, C, and threosidine. Instruments and methods useful in Raman spectroscopy like that in the present invention have been described for other applications. See, e.g., Toshima et al., Jpn J Ophthalmol, 1990; Nie et al., Exp Eye Res, 1990 (evaluation of lens water content, and cataract progression); Sebag et al., Invest Ophthalmol Visual Sci, 1994 (evaluation of the progression of retinopathy); Shim and Wilson, J Raman Spectroscopy, 1996 (characterization of fundamental Raman-active bonds in skin); Caspers et al., Biospectoscopy, 1998 and Caspers et al., J Invest Derm, 2001, (study of natural moisturizing factor in stratum corneum); Caspers et al., Biophysical J, 2002 (Raman confocal microscopy of skin).
The relatively long half-life (t1/2≅15 yr) of skin collagen and its associated AGEs make these species potential indicators of cumulative tissue glycemia. Levels of specific skin AGEs are correlated with the presence and severity of end-organ diabetes complications such as joint stiffness, retinopathy, nephropathy, and arterial stiffness. See Buckingham, Diabetes Care, 1984; Buckingham J Clin Invest, 1990; Monnier, NEJM 1986; Monnier, J Clin Invest 1986; Sell, Diabetes, 1992. In the largest such study to date, the DCCT Skin Collagen Ancillary Study Group evaluated a number of skin collagen variables from punch biopsies that were donated by a large fraction of the study's participants. These researchers found that skin AGEs were significantly correlated with the presence and clinical grade of diabetic neuropathy, nephropathy, and retinopathy. See Monnier et al., Diabetes, 1999.
The present invention can determine the diabetic state of a subject using one or more noninvasive Raman measurements. The invention can illuminate a portion of the tissue of the individual (e.g., a portion of the skin) with excitation light and detect Raman emission from the tissue. The characteristics of the Raman emission convey information about the disease state of the tissue under interrogation. The invention can apply additional processing algorithms to the measured Raman spectra before imposing a simple numerical threshold or a more detailed mathematical model to relate the optical information to disease state. In other embodiments, the output of the thresholding process or mathematical model can be a quantitative measure of diabetes-induced chemical change in the tissue of the individual being measured, rendered without regard to the individual's diabetic status. In additional embodiments, the invention can utilize a quantitative measure of diabetes-induced chemical changes in order to further infer or classify the diabetic status of the individual undergoing measurement.
Determining a Raman Emission of Tissue
Raman emission occurs when tissue inelastically scatters incident light. The wavelength difference between the incident and scattered light corresponds to the energy contributed to the vibrational energy of molecular constituents of the tissue. A Raman spectrum is the aggregate of wavelength-shifted, inelastically scattered radiation owing the specific structure and peaks to the chemical makeup of the tissue. Raman spectra convey specific information about the vibrational, stretching, and breathing bond energies of the illuminated sample. This molecular specificity provides insight into tissue composition and health. Detection of spectra features relating to advanced glycation endproducts is fundamental to screening and/or monitoring glucose-induced damage in the skin. While Raman spectroscopy is rich in information about the molecular composition of the tissue, it is minute in comparison to other optical phenomena such as fluorescence and Rayleigh (elastic) scattering. Line rejection filters (or similar) can be used to reduce interference from scattered incident light. Longer wavelength excitation sources can aid in reducing tissue fluorescence. Near infrared (NIR) excitation sources have the additional benefit of deeper penetration in tissue. Nevertheless, even using NIR excitation, tissue auto-fluorescence can be orders of magnitude more intense than the Raman scattering signal. Wavelength shift techniques or polynomial subtraction can be used to extract the comparatively weak Raman signal from the large, broad fluorescence spectra. See Shim and Wilson, J Raman Spectroscopy, 1997 and references cited therein.
When excitation light is launched into the tissue, it is subject to scattering and absorption processes that vary with the optical properties of the site under interrogation, the excitation wavelength, and the optical probe geometry. Emitted Raman light is also subject to wavelength- and location-dependent absorption and scattering as it propagates through the tissue prior to emergence and collection. Often, the tissue property of interest is its ‘intrinsic’ Raman emission, defined as the Raman emission that is scattered by a specimen that is homogeneous, nonscattering, and optically dilute. In order to accurately characterize the intrinsic Raman spectrum of the tissue of interest, the spectra-altering effects of scattering and absorption that are impressed upon the excitation and emitted light can be removed. Variations due to subject-to-subject and site-to-site differences can overwhelm the subtle spectral variations indicative of tissue status. Spectral correction based upon the tissue optics of each subject (at the same site as the Raman measurement, or at a different site having a predictable relationship to the site) can reveal the intrinsic Raman spectra of the molecules of interest. This intrinsic correction mitigates the variations across and within subjects, unmasking the spectral features relating to presence and state of disease.
The data described in this example were collected with a research Raman instrument based upon a 785 nm diode laser, a holographic transmission spectrometer and a CCD array detector. Residual fluorescence was removed by fitting a fifth-order polynomial to each Raman spectra. The polynomial acts as a wavelength-dependent baseline that is then subtracted to remove the predominant fluorescence.
Typical measured Raman spectra of skin are shown in
The optimum values for n and k are dependent on the arrangement of source and detector fibers, and can be determined empirically.
If multiple spectra are collected from each sample at an acquisition session, then the spectroscopic insertion variation, Sinsert, of the ith spectrum for specimen j can be expressed as the absolute deviation of that spectrum from the specimen's median:
S insert i,j(λ,n,k)=abs[E corr
An aggregate measure of insertion variation is then the variance of Sinsert:
νinsert(λ,n,k)=var(S insert(λ,n,k)). Eq 5
A variety of other procedures can accomplish intrinsic spectral correction. For example, a number of methods by which the measured spectra can be corrected using knowledge of the measured reflectance, tissue optical properties, and probe-dependent parameters can be used with the present invention. See, e.g., Gardner et al., Appl Opt, 1996, Zhang et al., Opt Lett, 2000; Muller et al., Appl Opt, 2001. In addition, intrinsic corrections can be made using a procedure in which the correction parameters for a given probe are created by measuring one or more tissue phantoms for which the emission, absorption, and scattering properties have been well characterized. This procedure can also be accomplished via Monte-Carlo or other computer simulation of the optical probe's response to media with known optical properties. Any of these processes can be used to correct for the effects of tissue optical properties in noninvasive skin Raman emission measurements.
While the examples described here generally concern steady-state Raman measurements without regard to polarization, it is possible to apply these methods to other Raman spectroscopy measurement modalities. For example, Fourier-transform Raman spectroscopy (FT Raman) in which the Raman emission is detected interferometrically can be suitable. Another suitable approach involves time-resolved techniques, in which a short burst of excitation light is applied to the tissue, and the Raman emission is time-gated as means to reject the ‘slower’ fluorescence. In addition, any of these techniques can be used in conjunction with an imaging methodology such as microscopy or macroscopic scanning of the excitation beam in order to acquire information about the spatial information. Any of the above-mentioned methods can be used in conjunction with a measurement technique that allows depth discrimination, such as a confocal detection system or optical coherence tomography, to add information concerning the distribution of Raman-active constituents with respect to depth beneath the tissue surface. The preceding discussion primarily describes detection of spontaneous Raman scattering as the means for gathering noninvasive, spectroscopic information relating to the disease state of the tissue. Other Raman techniques can also be incorporated in the invention. Examples of these alternative techniques are resonance Raman spectroscopy, surface-enhanced Raman scattering (SERS), anti-Stokes Raman scattering and coherent anti-Stokes Raman spectroscopy (CARS).
Determining a Model Relating Raman Properties to Disease State or Chemical Changes
The relationship between tissue Raman spectral properties at one or more wavelengths and diabetes disease state is typically not apparent upon visual inspection of the spectral data. Because this is the case, it is usually necessary that a multivariate mathematical relationship, or ‘model’, be constructed to classify tissue disease states or to quantify chemical changes using intrinsic Raman spectra. The construction of such a model generally occurs in two phases: (i) collection of ‘calibration’ or ‘training’ data, and (ii) establishing a mathematical relationship between the training data and the disease states or reference concentrations represented in the training data.
During the collection of training data, it can be desirable to collect Raman data from many individuals, representing disease states or reference values one wishes to characterize with the model to be constructed. For example, if one wishes to construct a model that separates diabetics from nondiabetics, it can be desirable to collect representative spectra from a wide variety of both types of individuals. It can be important to collect these data in a manner that minimizes the correlation between disease state and other parameters that can result in spectral variation. For example, the natural formation of collagen AGEs in health results in a correlation between skin AGE content and chronological age. It can be important, therefore, to obtain spectra from diabetics and nondiabetics spanning the ages for which the classification model is desired to be applicable. Alternatively, if one wished to construct a model that quantified the level of a specific skin collagen AGE, it can be advisable to collect spectroscopic data spanning a wide range of AGE reference values each day rather than to measure all individuals having the smallest AGE concentrations early in the study and all individuals with larger AGE concentrations later in the study. In the latter case, a spurious correlation arises between AGE concentration and time, and if there are instrumental trends over the course of the study, the resulting model might be calibrated to instrument state rather than analyte concentration.
As the training data are collected, additional reference information can be collected in order to later construct an appropriate classification model. For example, if the classification model is to predict diabetic state, the diabetes status of some or all of the individuals represented in the training set can be collected and associated with the corresponding spectroscopic training data. Alternatively, the classification model can predict the level of a certain chemical species in the skin, such as glycated collagen, glycated elastin, a specific AGE such as pentosidine or CML, or other proteins modified by the hyperglycemic conditions associated with diabetes mellitus. In these cases, skin biopsy specimens can be collected from individuals during the collection of training data. In addition, if other ancillary information, such as age, body mass index, blood pressure, HbA1c, etc. is to be used in generating later disease state assessments, this information can be collected for some or all spectra in the training set.
After the training data are collected, a multivariate model can be constructed to relate the disease states associated with the training data to the corresponding spectroscopic information. The exact model can be chosen based upon the ultimate goal of the training phase. There are at least two types of multivariate models that one might construct. In the first, the goal of the training process is to create a model that correctly classifies the disease state of the measured tissue. In this case, the output of the model is an assignment to one or more discrete classes or groups. These classes or groups might represent different grades or manifestations of a particular disease. They might also represent various degrees of risk for contracting a particular disease or other subgroups of the population that are pertinent to the disease state in question. For the second model type, the goal is to provide a quantitative estimate of some diabetes-induced chemical change in the system. The output of this model can be continuously variable across the relevant range of variation and is not necessarily indicative of disease status.
Classification of Tissue Disease Status
Model-building steps that can be followed when the end goal is to use the model to assess tissue disease state are depicted diagrammatically in
Finally, a technique for separating the data into the various classes can be selected. A variety of algorithms can be suitable, and the optimum algorithm can be selected according to the structure of the training data. In Linear Discriminant Analysis (LDA), a single linear function that best separates the multidimensional spectroscopic data into the reference classes observed in the training period is constructed. In Quadratic Discriminants Analysis, a quadratic discriminant function is constructed. FIG. 4 illustrates the manner in which the discriminant function might find the best separation between two groups—it depends on the structure of the data. In some cases (
A number of suitable classification algorithms exist. For example, k-nearest neighbors, logistic regression, hierarchical clustering algorithms such as Classification and Regression Trees (CART), and machine learning techniques such as neural networks, can all be useful and appropriate techniques. See, e.g., Huberty, Applied Discriminant Anaylsis, Wiley & Sons, 1994 and Duda, Hart, and Stork, Pattern Classification, Wiley & Sons, 2001.
Quantitation of Diabetes-induced Chemical Modifications
If the end goal is to quantify the concentration of an analyte or a class of analytes embedded in the tissue, a different approach can be taken in the model-building process. In this case, a set of (typically continuous) reference values for the analyte(s) in question can be obtained for some or all spectra in the training set. For example, in the event that the model is to quantify the level of CML in skin collagen, the reference concentrations associated with each spectrum in the training set can come from CML assays conducted on skin punch biopsy specimens obtained during calibration. In the event that the biopsy process is too invasive for the study participants, some surrogate for AGE-related chemical changes can also be used. For example, under the assumption that FPG values increase as the degree of diabetes progression increases, a reasonable compromise can collect FPG data as a surrogate for skin AGE concentration. HbA1c and OGTT information can be used similarly.
Calibration models used to predict quantitative values associated with a test set can be constructed by forming a mathematical relation between reference values and associated spectral data. A variety of algorithms are suitable. For example, in Principal Components Regression (PCR) the calibration data are first decomposed into a set of orthogonal scores and loadings, and then the reference values are regressed onto the scores of the first N PCA factors. Another suitable method is Partial Least-Squares (PLS) regression, in which a set of factors is constructed so that the squared covariance between the reference values and the scores on each successive PLS loading vector is maximized. These procedures and others have been summarized by Martens and Naes in Multivariate Calibration, Wiley & Sons (1989).
Quantitative calibration models other than the regression techniques described here can also be suitable for use with the present invention. Those skilled in the art will recognize that a variety of other approaches are available, including neural networks and other nonlinear techniques.
Determining Disease State or Chemical Changes from Raman Spectroscopy
After model construction, Raman spectral measurements can be made on new specimens having an unknown disease state or diabetes-related chemical change. The method by which the disease state or chemical properties of the new specimen are determined can depend on the type of model constructed in the training phase.
Classification of Tissue Disease Status
As mentioned above, a variety of models are available for discrimination of various diabetic states from measured Raman spectra. For example, when the method of Quadratic Discriminants Analysis is used, the new Raman spectrum is projected onto the factors created with the training data during construction of the classification model, creating a new vector of scores, xi, for the test spectrum. The means
D i,j=(x i −
The posterior probability that test sample i is a member of class j, p(iεj), can be calculated using Equation 8. As with all probabilities, this number ranges between 0 and 1; probabilities close to 1 indicate that an observation lies close to the diabetic class, and probabilities close to 0 indicate that an observation lies close to the non-diabetic class. The probability that sample i is a member of class j is given by
where πij are the prior probabilities that test sample i is a member of class j based on other knowledge (risk factors, etc.). The prior probabilities are parameters that can be tuned in the prediction phase depending, in part, on the diagnostic application of the classification algorithm.
Finally, a threshold can be applied that assigns the new Raman spectral measurement to a particular tissue disease state. For example, it might be determined that all Raman measurements yielding a posterior probability of diabetes greater than 0.75 will be assigned to the diabetic class. Like the prior probabilities, the exact threshold applied in validation can depend on a variety of factors, including the application, disease prevalence, and socioeconomic ramifications of positive and negative test results.
Quantitation of Diabetes-Induced Chemical Modifications
The output of a quantitative calibration model can be a regression vector that converts the corrected Raman spectrum into a quantitative analyte prediction via an inner product:
â=F corr ∘b, Eq 9
where â is the analyte prediction and b is the regression vector.
The method for generating a quantitative output can vary with the model constructed in the training phase. Final analyte quantitation with, for example, a neural network proceeds by a different process but yields a similar output.
After the construction of either type of multivariate model (i.e., a quantitative model for chemical change or a classification model for tissue disease state), the accuracy of the model can be tested by predicting the disease status associated with well-characterized ‘validation’ spectra. A variety of techniques also exists for accomplishing this task. In leave-one-out cross-validation, a single spectrum or set of spectra from the training set are omitted from the model-building process, and then the resulting model is used to predict the disease status associated with the spectra left out of the model. By repeating this process a sufficient number of times, it is possible to develop a mathematical estimate of the performance of the model under new conditions. A more rigorous test of the newly constructed model is to apply the model to an entirely new data set, or a ‘test’ set. In this case, the disease status associated with each spectrum is known, but the ‘test’ spectra are collected at a different time (e.g., subsequent to model-building) than the training data. By comparing the predictions on the ‘test’ data to the reference values associated with these data, the diagnostic accuracy of the model in question can be assessed independent of the training data.
Components or sub-systems of an apparatus to characterize and/or quantify disease state by tissue Raman emission are illustrated in
Example I of such a system embodies a continuous-wave diode laser as the light source. The optical-coupling sub-system is comprised of a fiber probe that couples the excitation light to the tissue and collects Raman scattering emanating from the tissue, as illustrated in
In Example II, the fiber-optic probe of Example I is replaced by a system of lenses and mirrors to convey excitation light from the light source to the tissue and then collect Raman emission from the tissue and relay it to the detection sub-system.
In Example II, the continuous-wave laser is replaced by a pulsed, modulated or mode-locked laser.
In Example III, the diode laser is replaced by a solid-state or gas laser.
In Example V, the detection system of Example I comprised of a spectrograph and an array detector is replaced by a monochromator and a photomultiplier.
An example of a fixture—in this instance, a forearm cradle—to hold the fiber bundle in contact with the skin of the subject is shown in
Those skilled in the art will recognize that the present invention can be manifested in a variety of forms other than the specific embodiments described and contemplated herein. Accordingly, departures in form and detail can be made without departing from the scope and spirit of the present invention as described in the appended claims.
|Cited Patent||Filing date||Publication date||Applicant||Title|
|US5313941||Jan 28, 1993||May 24, 1994||Braig James R||Noninvasive pulsed infrared spectrophotometer|
|US5348003||Sep 3, 1992||Sep 20, 1994||Sirraya, Inc.||Method and apparatus for chemical analysis|
|US5553616 *||Nov 30, 1993||Sep 10, 1996||Florida Institute Of Technology||Determination of concentrations of biological substances using raman spectroscopy and artificial neural network discriminator|
|US5582168||Jan 22, 1993||Dec 10, 1996||Georgia Tech Research Corp.||Apparatus and methods for measuring characteristics of biological tissues and similar materials|
|US5612540||Mar 31, 1995||Mar 18, 1997||Board Of Regents, The University Of Texas Systems||Optical method for the detection of cervical neoplasias using fluorescence spectroscopy|
|US5615673 *||Mar 27, 1995||Apr 1, 1997||Massachusetts Institute Of Technology||Apparatus and methods of raman spectroscopy for analysis of blood gases and analytes|
|US5652653||Aug 26, 1996||Jul 29, 1997||Eastman Chemical Company||On-line quantitative analysis of chemical compositions by Raman spectrometry|
|US5784162||Dec 12, 1995||Jul 21, 1998||Applied Spectral Imaging Ltd.||Spectral bio-imaging methods for biological research, medical diagnostics and therapy|
|US5845639 *||Nov 11, 1994||Dec 8, 1998||Board Of Regents Of The University Of Washington||Optical imaging methods|
|US5855882||Jun 7, 1995||Jan 5, 1999||The Picower Institute For Medical Research||Methods for inhibiting the cross-linking of advanced glycosylation endproducts|
|US5861238||Jun 7, 1995||Jan 19, 1999||The Picower Institute For Medical Research||Methods for partitioning advanced glycosylation endproducts|
|US5864397||Sep 15, 1997||Jan 26, 1999||Lockheed Martin Energy Research Corporation||Surface-enhanced raman medical probes and system for disease diagnosis and drug testing|
|US5882301||Dec 12, 1996||Mar 16, 1999||Yoshida; Akitoshi||Measuring apparatus for intraocular substance employing light from eyeball|
|US5885224||Oct 30, 1996||Mar 23, 1999||Yoshida; Akitoshi||Intraocular substance measuring apparatus|
|US5962245 *||Jun 7, 1995||Oct 5, 1999||The Picower Institute For Medical Research||Methods for detecting the presence of advanced glycosylation endproducts|
|US6044285||Nov 12, 1998||Mar 28, 2000||Lightouch Medical, Inc.||Method for non-invasive measurement of an analyte|
|US6167290||Feb 3, 1999||Dec 26, 2000||Bayspec, Inc.||Method and apparatus of non-invasive measurement of human/animal blood glucose and other metabolites|
|US6289230||Jul 7, 1999||Sep 11, 2001||Lightouch Medical, Inc.||Tissue modulation process for quantitative noninvasive in vivo spectroscopic analysis of tissues|
|US6352502||Dec 3, 1999||Mar 5, 2002||Lightouch Medical, Inc.||Methods for obtaining enhanced spectroscopic information from living tissue, noninvasive assessment of skin condition and detection of skin abnormalities|
|US6353226||Nov 23, 1998||Mar 5, 2002||Abbott Laboratories||Non-invasive sensor capable of determining optical parameters in a sample having multiple layers|
|US6377828||Jan 10, 2000||Apr 23, 2002||Lightouch Medical, Inc.||Method for non-invasive measurement of an analyte|
|US6411838 *||Dec 22, 1999||Jun 25, 2002||Medispectra, Inc.||Systems and methods for optical examination of samples|
|US6485703||Jul 9, 1999||Nov 26, 2002||The Texas A&M University System||Compositions and methods for analyte detection|
|US6505059||Apr 6, 1999||Jan 7, 2003||The General Hospital Corporation||Non-invasive tissue glucose level monitoring|
|US6522903||Oct 19, 2000||Feb 18, 2003||Medoptix, Inc.||Glucose measurement utilizing non-invasive assessment methods|
|US6560478||Nov 20, 2000||May 6, 2003||The Research Foundation Of City University Of New York||Method and system for examining biological materials using low power CW excitation Raman spectroscopy|
|US6571117 *||Aug 10, 2001||May 27, 2003||Ralf Marbach||Capillary sweet spot imaging for improving the tracking accuracy and SNR of noninvasive blood analysis methods|
|US6574490||Apr 11, 2001||Jun 3, 2003||Rio Grande Medical Technologies, Inc.||System for non-invasive measurement of glucose in humans|
|US6574501||Jan 9, 2001||Jun 3, 2003||Childrens Hospital Los Angeles||Assessing blood brain barrier dynamics or identifying or measuring selected substances or toxins in a subject by analyzing Raman spectrum signals of selected regions in the eye|
|US6609015 *||Jul 24, 2001||Aug 19, 2003||Koninklijke Philips Electronics N.V.||Analysis of a composition|
|US6615061||Aug 3, 1999||Sep 2, 2003||Abbott Laboratories||Optical sensor having a selectable sampling distance for determination of analytes|
|US6630673||Dec 21, 2001||Oct 7, 2003||Abbott Laboratories||Non-invasive sensor capable of determining optical parameters in a sample having multiple layers|
|US6662030 *||May 18, 1998||Dec 9, 2003||Abbott Laboratories||Non-invasive sensor having controllable temperature feature|
|US6665556 *||Jan 28, 2000||Dec 16, 2003||Robert R. Alfano||Method and apparatus for examining a tissue using the spectral wing emission therefrom induced by visible to infrared photoexcitation|
|US6721583 *||Nov 18, 1999||Apr 13, 2004||The United States Of America||Method for non-invasive identification of individuals at risk for diabetes|
|US6760613 *||Jun 24, 2002||Jul 6, 2004||Medispectra, Inc.||Substantially monostatic, substantially confocal optical systems for examination of samples|
|US6853854 *||Jan 21, 2000||Feb 8, 2005||Q Step Technologies, Llc||Noninvasive measurement system|
|US6889075 *||May 3, 2001||May 3, 2005||Rocky Mountain Biosystems, Inc.||Optical imaging of subsurface anatomical structures and biomolecules|
|US7113814 *||Jul 13, 2001||Sep 26, 2006||Virginia Commonwealth University||Tissue interrogation spectroscopy|
|US20020091324 *||Feb 28, 2002||Jul 11, 2002||Nikiforos Kollias||Non-invasive tissue glucose level monitoring|
|US20040019283 *||Mar 25, 2003||Jan 29, 2004||Lambert James L.||Assessing blood brain barrier dynamics or identifying or measuring selected substances, including ethanol or toxins, in a subject by analyzing Raman spectrum signals|
|US20040092590 *||Sep 29, 2003||May 13, 2004||Linda Arterburn||Glycemic control for prediabetes and/or diabetes Type II using docosahexaenoic acid|
|Citing Patent||Filing date||Publication date||Applicant||Title|
|US8125623 *||Mar 30, 2009||Feb 28, 2012||Ottawa Hospital Research Institute||Correlation technique for analysis of clinical condition|
|US20110313296 *||Dec 22, 2011||Johnson Robert D||Method and Apparatus for Determination of a Measure of a Glycation End-Product or Disease State Using Tissue Fluorescence|
|U.S. Classification||600/310, 600/342, 600/478, 600/407, 398/79, 600/476, 356/301|
|International Classification||A61B5/103, A61B5/00|
|Cooperative Classification||A61B5/061, A61B5/0071, A61B5/441, A61B5/0068, A61B5/0066, A61B5/0064, A61B5/412, A61B5/443, A61B5/0059, A61B5/0075|
|European Classification||A61B5/06C, A61B5/44B, A61B5/00P1A, A61B5/00P1C, A61B5/00P1E, A61B5/00P5, A61B5/00P7, A61B5/44B3, A61B5/41D, A61B5/00P|
|Feb 2, 2004||AS||Assignment|
Owner name: INLIGHT SOLUTIONS, INC., NEW MEXICO
Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNORS:EDIGER, MARWOOD NEAL;HULL, EDWARD L.;GARDNER, CRAIG M.;REEL/FRAME:015057/0813;SIGNING DATES FROM 20031217 TO 20040128
Owner name: INLIGHT SOLUTIONS, INC.,NEW MEXICO
Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNORS:EDIGER, MARWOOD NEAL;HULL, EDWARD L.;GARDNER, CRAIG M.;SIGNING DATES FROM 20031217 TO 20040128;REEL/FRAME:015057/0813
|Jul 12, 2006||AS||Assignment|
Owner name: VERALIGHT, INC.,NEW MEXICO
Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNOR:INLIGHT SOLUTIONS, INC.;REEL/FRAME:017921/0693
Effective date: 20060710
|Nov 27, 2013||FPAY||Fee payment|
Year of fee payment: 4
|Nov 27, 2013||SULP||Surcharge for late payment|